List your company
ResearchFormerly WizWand

Open-ended generation on LLaVA-Bench In-the-Wild

109.3ScoreBaseline
73.9483.1292.3101.48May 19, 2026
Updated 5mo ago

Evaluation results

MethodLinks
BaselineBase Model=QWEN2.5-VL-7B-INSTRUCT2026.05109.3———
FastV + RotateKBase Model=QWEN2.5-VL-7B-INSTRUCT, Token Sparsity Ratio=0.30, Channel Sparsity Ratio=0.252026.05106.9———
VisionZip + RotateKBase Model=QWEN2.5-VL-7B-INSTRUCT, Token Sparsity Ratio=0.35, Channel Sparsity Ratio=0.252026.05106.2———
VisionZipBase Model=QWEN2.5-VL-7B-INSTRUCT, Token Sparsity Ratio=0.222026.05104———
FastVBase Model=QWEN2.5-VL-7B-INSTRUCT, Token Sparsity Ratio=0.18752026.05100———
FastV + ThinKBase Model=QWEN2.5-VL-7B-INSTRUCT, Token Sparsity Ratio=0.30, Channel Sparsity Ratio=0.252026.0599.8———
VisionZip + ThinKBase Model=QWEN2.5-VL-7B-INSTRUCT, Token Sparsity Ratio=0.35, Channel Sparsity Ratio=0.252026.0599.1———
FastV + RotateKBase Model=LLAMA3-LLAVA-NEXT-8B, Token Sparsity Ratio=0.30, Channel Sparsity Ratio=0.252026.0585.5———
VisionZip + RotateKBase Model=LLAMA3-LLAVA-NEXT-8B, Token Sparsity Ratio=0.35, Channel Sparsity Ratio=0.252026.0584.7———
FastVBase Model=LLAMA3-LLAVA-NEXT-8B, Token Sparsity Ratio=0.18752026.0584.6———
BaselineBase Model=LLAMA3-LLAVA-NEXT-8B2026.0584.1———
VisionZipBase Model=LLAMA3-LLAVA-NEXT-8B, Token Sparsity Ratio=0.222026.0581.1———
FastV + ThinKBase Model=LLAMA3-LLAVA-NEXT-8B, Token Sparsity Ratio=0.30, Channel Sparsity Ratio=0.252026.0580.4———
VisionZip + ThinKBase Model=LLAMA3-LLAVA-NEXT-8B, Token Sparsity Ratio=0.35, Channel Sparsity Ratio=0.252026.0575.3———
BEAM2025.03—58.9474.7178.25
Greedy2025.03—56.9471.8379.95
IBD2025.03—61.5977.9182.86
ICD2025.03—54.4366.8974.81
M3ID2025.03—61.8877.4982.83
OPERA2025.03—58.4471.8177.92
PAI2025.03—52.2969.7472.49
PM2025.03—62.4677.2184.5
SID2025.03—59.4970.9381.63
VCD2025.03—57.1769.481.22
VDD2025.03—61.678.2283.01